An improved exemplar-based image inpainting algorithm
Chunyang Xiang, Yangjie Cao, Pengsong Duan, Lei Shi · 2014
Aiming to address deficiencies existing in traditional image inpainting algorithms, an improved image inpainting algorithm, named S-Criminisi, is proposed. Compared with traditional image inpainting algorithms, advantages of S-Criminisi are: (1) Instead of using traditional curve fitting method, S-Criminisi utilizes the Difference Method calculating gradient operator and obtaining curvature of target point; (2) To solve discontinuities of visual images caused by error matching, a new matching method based on matrix similarity is applied by S-Criminisi. Experimental results show that S-Criminisi is more suited for characteristics of digital image and be a better solution for discontinuities of visual images.